I recently came across an ML method called ARIMA (Autoregressive Integrated Moving Average)+GARCH(Generalized autoregressive conditional heteroscedasticity).
To summarize, ARIMA uses historical time series data to predict future points in data. GARCH is similar to ARIMA except that it is used to predict error variances in the data (volatility). They can be used together to estimate data in relation to historical volatility and even seasonal market regimes. There is a file on the MATLAB file exchange which shows a good example of it being used to forecast long term energy prices based off years of data. The model's prediction was compared to out of sample data and it did amazingly well.
Has anyone here used and implemented this method into their trading? If so, do you use it to complement existing trading strategies or do you use it to assist in finding edges in your data? Did you find it useful? How does it compare to other ML predictive methods out there? I would love to see a discussion about this.